Transitive Inference in Rats Using
Odor Stimuli: Manual Versus
Automated Training Procedures
Mary Elizabeth Pacewicz
University of North CarolinaWilmington
Faculty Mentor: Kate Bruce
University of North CarolinaWilmington
ABSTRACT
Transitivity is a type of higher-order learning that is demonstrated when untrained relations emerge
after specific relations have been trained. The formation of a hierarchy is required to determine the
transitive relations between stimuli. To train the hierarchy A<B<C<D<E, the pairs A-B+, B-C+,
C-D+, and D-E+ are presented (the plus sign indicates the reinforced stimulus). After training, novel
pairs, such as A-C+ and B-D+ are presented to test for transitive inference. This study compares three
rats’ performances on tests of novel pairs in a manual procedure and four rats’ performances in an
automated procedure. While automated procedures generally increase efficiency and objectivity and
are a preferred method to manual procedures, rats trained using the automated apparatus failed to
meet baseline training criterion and were not successful at tests of novel pairs. All rats trained with
the manual procedure demonstrated transitive inference. The ecological significance of transitive
inference and the testing procedure is discussed.
INTRODUCTION
oncept learning is a type of higher-order learning requiring the formation
of categories. Categories are formed by
utilizing features, such as shape or function,
to group stimuli (Lazareva & Wasserman,
2008). Perceptual concepts, nonsimilaritybased concepts, and abstract concepts are
three types of concepts used to evaluate
and categorize stimuli. Perceptual concepts are shown when physical dimensions
of the stimuli are the basis for categorization. Nonsimilarity-based concepts include
stimuli that are physically different but have
a similar function (Lazareva & Wasserman,
2008).
Abstract concepts are demonstrated
when the relations among the stimuli are
used to categorize the stimuli in context
C
with one another (Lazareva & Wasserman,
2008). For example, if it has been trained
that stimulus A equals stimulus B (A=B)
and stimulus B equals stimulus C (B=C),
the presentation of stimulus A and C would
create a novel situation. However, by utilizing information from previously trained
pairs, a relation between the two could be
made. This relation would be that stimulus A equals stimulus C (A=C) (Sidman,
Rauzin, Lazar, Cunningham, Tailby, &
Carrigan, 1982). The example of the relation between stimulus A and C is an example of transitivity, one of the properties
of the equivalence concept.
Equivalence involves making relations about items that have never been
trained together (Davis, 1992; Lazareva
& Wasserman, 2006). For example, if the
135
Explorations | Social Sciences
pairs A=B and B=C are trained during
baseline training, they are grouped into one
class. A class is a group of specific stimuli
that are interchangeable (McIntire, Cleary,
& Thompson, 1987). Training that classifies stimuli A, B, and C as one class allows
novel probe tests to be given to test the relations that have been formed between the
stimuli. The emergence of untrained relations between novel pairs such as reflexivity (A=A, B=B, C=C), symmetry (B=A,
C=B), and transitivity (A=C, C=A) represents equivalence; the stimuli have never
been presented together or in that order,
yet subjects may respond as if the stimuli
are related (Sidman, 2000). Nonhumans
do not typically show the formation of all
equivalence relations. In fact, the formation of equivalence relations has been considered a form of higher-order learning
that requires the use of language (Hayes,
1989). However, Schusterman and Kastak
(1993) provided the first concrete evidence
that, given extensive training, a nonhuman
animal (a sea lion) could form all equivalence relations.
Schusterman and Kastak (1993) trained
a California sea lion on 30 equivalence
classes in order to provide multiple exemplars. Each class was composed of three
stimuli (A, B, and C). The first phase of
training involved presentations of stimuli
A1-A30 as samples and B1-B30 as comparisons. An example of this training is presenting A1 as the sample and B1 and BX as
comparisons. Responses to B1 were reinforced and responses to any other B stimulus (BX) would not result in reinforcement.
After this training, 12 of the A=B pairs
were tested and then trained for symmetry
(B=A). Not until the sea lion performed at
a high level on presentations of B=A did
the training of B=C begin. Once the sea
lion met criterion on the B=A training, 30
B=C pairs were trained. Then, the same
12 pairs that were used to test B=A were
used to test and train symmetry with the
136
B=C pairs. After this symmetry training
(C=B), the same 12 potential equivalence
classes were tested and trained for transitivity (A=C). Following this training, those
12potentialequivalenceclassesweretested
for C=A symmetry. This extensive training provided multiple exemplars for the sea
lion before the final test of equivalence was
conducted.
A higher-order learning task that involves the presentation of novel pairs of
stimuli to determine the relations that
have been made is a test for transitive inference. In a typical procedure to test for
transitive inference, a hierarchical order of
the stimuli is trained. For example, given
the two statements “Alex is shorter than
Brad” and “Brad is shorter than Chris,”
the hierarchy in terms of height would
be Alex<Brad<Chris. The inference that
Alex is shorter than Chris can be determined due to their relationships to Brad in
the hierarchy.
The training of a hierarchy is done by
presenting pairs of stimuli and always reinforcing responses to one stimulus and never
reinforcing responses to the other stimulus
(Lazareva et al., 2004). For example, given
the three stimuli A, B, and C, the hierarchical order A<B<C can be trained. This
can be done by presenting two pairs, A-B
and B-C. In the pair A-B, responses to B
are always reinforced and responses to A
are never reinforced. With the pair B-C,
responses to C are always reinforced and
responses to B are never reinforced. Once
the training of the hierarchical order is
complete, the test for transitive inference
can be presented. This involves the presentation of the novel pair A-C. Using the
trained hierarchy, a relation between stimulus A and stimulus C can be made (A<C).
Though information relating stimulus A
and C has not been directly trained, a relation between the two stimuli can be inferred due to their relation to stimulus B
(A<B<C) (Davis, 1992).
Mary Beth Pacewicz
Though a hierarchical order can be established using three stimuli, the novel
pair (A-C) does not represent a true test of
transitive inference. Stimuli A and C are
called end stimuli, and success on a novel
pair containing end stimuli can be explained by reinforcement histories (Dusek
& Eichenbaum, 1997). Therefore, a minimum of a five-item list is needed to ensure
that the novel test pair does not include an
end stimulus (Weaver, Steirn, & Zentall,
1997). In a five-item list (A, B, C, D, and
E), training would occur by presenting the
following pairs: A-B, B-C, C-D, and D-E.
Within these pairs, one stimulus would
always be reinforced and the other would
never be reinforced, A-B+, B-C+, C-D+,
and D-E+ (the plus sign indicates the reinforced stimulus).
The five-item list allows for a true test
of transitive inference because two of the
middle stimuli have equal reinforcement
histories and have never been presented together (stimulus B and stimulus D). Success
on B-D+ demonstrates transitive inference
because both B and D have been previously
reinforced 50 percent of the time they were
presented and not reinforced the other 50
percent of the time. B and D have equal
reinforcement histories and therefore, during a transitive inference test, the choice of
D+ indicates that a relation between stimuli
has been formed (Weaver et al., 1997).
Transitive inference has been demonstrated in a number of non-human species including chimpanzees (Gillan, 1981),
rhesus macaques (Treichler & Van Tilburg,
1996), hooded crows (Lazareva et al.,
2004), pigeons (von Fersen, Wynne, Delius,
& Staddon, 1991; Lazareva & Wasserman,
2006), and rats (Davis, 1992; Dusek &
Eichenbaum, 1997). Davis (1992) and
Dusek and Eichenbaum (1997) used olfactory stimuli to test transitive inference in
rats. In both studies, rats have successfully
performed tasks using olfactory stimuli
rather than visual stimuli. Rats succeeded
in the transitive inference tasks even though
the olfactory stimuli were not orderable on
any dimension other than reinforcement
history.
Davis (1992) trained four rats with the
series A-B+, B-C+, C-D+, and D-E+ to
establish the hierarchy A<B<C<D<E.
Olfactory stimuli were used, and the apparatus was a “Y” maze with a door at the
end of each arm of the “Y.” The doors
were scented with different food flavoring.
The rats were trained on the following:
A-B+, B-C+, A-C+, C-D+, D-E+, C-E+, and
A-E+ . Then, the novel pair, B-D+ was presented, and the rats chose D over B even
without training.
Dusek and Eichenbaum (1997) also
used olfactory stimuli, cups of scented
sand, to train rats with a five item list
(A>B>C>D>E). Pairs were trained in five
phases similar to Davis (1992). Transitive
inference was tested with 10 trials of the
B-D pair and 10 trials of the A-E pair.
New scents (W-X and Y-Z) were also presented to measure the rate at which the rats
learned novel pairs. The rats demonstrated
transitive inference when tested on the B-D
pair and showed some evidence for learning the novel scent pairs.
Jordan (2009) replicated Dusek and
Eichenbaum’s (1997) methods to train three
rats on two five-item lists. She reported results from two rats and demonstrated transitive inference. In the current study, I will
present results from the third rat trained
with the Jordan (2009) procedure. Her procedure involved the manual presentation
of all stimuli, and I will also compare her
findings related to transitive inference using a manual procedure to those obtained
using an automated procedure.
Success with olfactory tasks has been
demonstrated with rats in automated apparatuses. Otto and Eichenbaum (1992)
trainedratsinadelayednon-match-to-sample task using odor stimuli that were presented through a nose poke in the operant
137
Explorations | Social Sciences
chamber. The rats rapidly learned the
task. Lionello-DeNolf and Mihalick (2006)
used an automated olfactory apparatus to
tests rats’ ability to learn a simple discrimination task, followed by a reversal. The results of the study supported the use of an
automated apparatus in olfactory tasks and
suggestedthatmorecomplextaskscouldbe
tested using an automated apparatus.
The use of an automated apparatus increases both efficiency and objectivity, and
therefore, is a preferred method to manual
procedures. Transitive inference tasks have
never been tested in an automated apparatus, and the objective of this study was
to train rats to form a hierarchy with five
scents and test the performance on a transitive inference task.
We expected that transitive inference
would be apparent using the manual and
automated apparatuses. We hypothesized
that rats’ performances on the B-D+ probe
would be above chance levels, and the rats
would choose the D stimulus over the B
stimulus. We also expected that rats would
learn the B-D+ relation more rapidly than a
relation between two unfamiliar, untrained
scents.
METHOD
Manual Procedure
Subjects
Three male Sprague-Dawley rats were
trainedusingamanualproceduredescribed
by Jordan (2009). Jordan reported data
from two rats (P39 and P23), and data from
rat S16 is reported here. Rats were kept
at 85-90% of their free-feeding weight with
unlimited access to water. Approximately
30 minutes after finishing their session,
the rats were fed 10-15 grams of Purina®
Rat Chow. They were housed individually in a colony room that was illuminated
on a 12- hour reversed light/dark cycle.
Apparatus
Testing occurred in an operant chamber
with a modified front wall. The wall had a
138
Figure 1. Manual apparatus. The operant chamber is
shown in top photograph. The rat making a response is
shown in lower photograph.
gap allowing a Plexiglas tray to be pushed
into the chamber to deliver the stimuli during testing (see Figure 1).
Stimuli were lids scented with spices purchased from the Great American and the
Rocky Mountain Spice Companies. Lids
were stored for at least 24 hours before use
in containers with spices (caraway, marjoram, celery, mustard, allspice, lime, anise,
sage, onion, paprika, raspberry, thyme,
cinnamon, garlic, savory, coriander, bay,
carob,Worcestershire, tomato, fennel, beet,
ginger, turmeric, cumin, nutmeg, clove,
and dill).
Scented lids were placed on top of
2-ounce plastic cups. A layer of sand was
placed in the bottom of each cup. Sucrose
pellets (.45 mg) were used as reinforcers
and were buried 0.5-0.8 millimeters below
the surface of the sand underneath the
scented lids. The cups with scented lids
on top were placed in two holes (5 cm diameter) cut in the Plexiglas tray. All housing and testing procedures were approved
Mary Beth Pacewicz
by the Institutional Animal Care and Use
Committee.
Procedure
Shaping
The rats were first trained to dig for
the sucrose pellets. The first step in this
training involved putting the sucrose pellets on top of the cups containing sand.
Eventually, the pellets were buried between 0.5- 0.8 mm below the surface.
The rats were then trained to remove lids
by pushing them off of the sand cups.
Five-Item List Training
Two five-item lists were trained as described in Jordan (2009). List one consisted
of scents A-E, and list two consisted of
scents F-J. Scents were presented in pairs
of two (A-B+, B-C+, C-D+, D-E+, F-G+,
G-H+, H-I+, and I-J+), to create a hierarchy (A<B<C<D<E and F<G<H<I<J).
In each pair a response to one scent was
always reinforced and response to the
other scent never reinforced. Pushing the
scented lid off of the cup was counted as
a response. Correct responses were reinforced with a sucrose pellet. If an incorrect
response was made, the Plexiglas tray was
immediately removed from the chamber.
If two minutes elapsed without a response
to either cup, the tray was removed from
the chamber and the next trial began.
The first five-item list was trained in four
phases. The first phase consisted of the
presentation of each pair in blocks of six
trials. Six trials of A-B+ were presented,
followed by B-C+, C-D+, and D-E+, respectively. No more than two identical trials
were presented in a row, and the reinforced
scents were equally balanced between the
left and right side. Criterion of 80% correct on each trial type for two consecutive
days was required before proceeding to the
following phase. In phase two, three trials
of A-B+ were presented followed by three
trials of B-C+, then C-D+, and D-E+. This
sequence was repeated for a total of 24 trials. Six blocks of four trials were presented
in phase three. Each block had one A-B+,
B-C+, C-D+, and D-E+ pair presented in a
random order. Phase four consisted of 24
trials with pairs presented in a random order. Six trials of each pair were presented
with no more than two identical trials occurring in a row.
Once phase four criterion was met, a
probe session followed. The session began with eight baseline pair trials. Two of
each pair were presented, and 7/8 correct
was required for the rat to be given novel
pairs. If 7/8 correct was not achieved, the
remainder of the session consisted of phase
four training. If the criterion of 7/8 was
reached, the probe test was given. Probe
testing included two presentations of A-C+,
A-D+, A-E+, B-E+, C-E+, B-D+, X-Y+, and
one trial of each baseline pair. The X-Y+
pair consisted of two novel scents, used
to control for learning during the probe
testing.
After completion of probe testing, two
more days of training were given in phase
five. Phase five included two trials of A-C+,
A-D+, A-E+, B-E+, C-E+, three trials of
B-D+ and X-Y+, and two trials of each baseline pair. Following the two days of phase
five, the second five-item list was trained.
The second five-item list was trained in the
same way the first list was trained.
Automated Procedure
Subjects
Four male Sprague-Dawley rats were
trained for the automated procedure. They
were kept at 85-90% of their free-feeding
weight and given unlimited access to water. They were fed about 10-15 grams of
Purina® Rat Chow after a minimum of 30
minutes after the completion of the testing
session. The rats were kept on a 12 hour
reverse light/dark cycle.
Apparatus
All housing and testing procedures were
approved by the Institutional Animal Care
and Use Committee. An operant chamber was used for training and testing (see
139
Explorations | Social Sciences
Figure 2. Automated apparatus. The operant chamber is shown in top left. The nose pokes on the left side of the
chamber are shown in lower left. The top right image shows the olfactometers (where scented oil was stored). The lower
right image shows the solenoids.
Figure 2A). The left side of the chamber
had three nose pokes, with photo beam
sensors that indicated when a response was
made (see Figure 2B). Nose pokes were
the response ports where scented air was
pumped in. Only the left and right nose
pokes were used in the transitive inference
procedure. The right side of the chamber
stored a pellet dispenser. Correct responses
were reinforced by the presentation of a sucrose pellet. A house light was used to indicate the beginning and end of the session
and to indicate a timeout after an incorrect
response.
The stimuli used were scented oils from
the Great American Spice Company.
Peach, grape, bubblegum, strawberry, and
banana were the scents used during baseline training and probe testing. Raspberry,
vanilla, blueberry, and pineapple were
used as unfamiliar scents during probe
testing. Five ml of each scent was poured
into each of three glass jars. One jar was
attached to each olfactometer (see Figure
2C). One olfactometer was assigned to
each nose poke. The jars were connected
140
to the olfactometers by a lid that was connected to a pump. There was one pump
per jar, and each pump was attached to a
solenoid (see Figure 2D). Solenoids could
be opened or closed, allowing for the corresponding scent to be pumped into the
nose poke. A vacuum pump was used to
constantly remove the scented air from the
nose pokes to ensure that it did not spread
into the chamber.
MedPC software was used to create the
transitive inference programs. Four different programs were used in each phase.
The programs were executed on a Dell
desktop computer which was connected to
the operant chamber.
Procedure
Magazine Training
One pellet was delivered and the hopper
light turned on. The hopper was where
the pellets were delivered, and the light was
turned on to cue the rats that reinforcement had been delivered. The hopper light
remained on for 45 seconds. After an intertrial interval of 15 seconds, another pellet
was delivered and the hopper light turned
Mary Beth Pacewicz
on. This procedure was repeated 25 times,
with a total of 25 pellets delivered. After
the completion of the session, the pellets
remaining in the hopper were counted.
Criterion to move to any nose poke shaping was the consumption of all 25 pellets.
Any Nose Poke Shaping
Sessions ran for 25 minutes. Both the left
and right nose pokes were activated, and a
response to either nose poke resulted in reinforcement. One pellet was delivered and
the hopper light remained on for five seconds. One disruption of the photo beam
was counted as one response. Criterion to
move to nose poke shaping was set at 25
responses in one session.
Nose Poke Shaping
Sessions consisted of 48 trials. For a
trial, either the left or right nose poke
was activated. A response to the activated nose poke resulted in reinforcement.
Reinforcement consisted of the delivery of
a pellet. The hopper light was turned on
and remained on for five seconds. A response was defined as breaking the photo
beam in the nose poke. For the first day of
nose poke shaping, only one response was
required to receive reinforcement (FR1).
Each day, the fixed ratio was increased by
one (e.g. FR2 on day two). If responding
slowed or stopped, the fixed ratio was decreased. Criterion was set at 45 rewards on
a fixed ratio of five responses.
Five-Item List Training
Rats were trained to form a hierarchy of
five scents (A, B, C, D, and E) by presenting the scents in four pairs (A-B+, B-C+,
C-D+, and D-E+). In each pair a response
to one scent was always reinforced and response to the other scent never reinforced.
Responseswerecountedwhentheratbroke
a photo beam in the nose poke. A correct
response resulted in the hopper light turning on for five seconds as a pellet was delivered. If an incorrect response was made,
the house light immediately turned off and
resulted in a five second timeout. After the
fivesecondtimeout,thenextsessionbegan.
This non-correction procedure was utilized
throughout all phases.
Each rat was trained with the five scents:
peach (A), grape (B), bubblegum (C), strawberry (D), and banana (E) during the 48
daily trials. Phase one consisted of four
blocks of 12 trials. Only the A-B+ pair
was presented in block one, followed by the
B-C+ pair in block two, the C-D+ pair in
block three, and the D-E+ pair in block four.
Correct responses were reinforced with one
sugar pellet which was released into the
hopper. The hopper light remained on for
five seconds. Incorrect responses resulted
in a five second timeout. Criterion to move
to each phase was set at 10/12 (83%) correct on each type of pair for two consecutive days. In phase two, each pair was presented in blocks of six. A-B+ was the first
block, followed by B-C+, C-D+, and D-E+
respectively. The blocks were then repeated
so a total of 12 trials of each pair was completed. Phase three consisted of blocks of
four trials. A-B+ was the first block of, then
B-C+, C-D+, and D-E+. Once four trials of
the D-E+ pair were completed, A-B+ pairs
were presented again, followed by B-C+,
etc. This order was repeated three times
for a total of 12 trials of each pair. In phase
four, each pair was presented once, starting
with A-B+ and ending with D-E+. This sequence was repeated twelve times. The final phase, phase five, consisted of blocks of
four trials in which each pair was presented
once but in a random order. For example,
the first block may be B-C+, A-B+, C-D+,
and then D-E+ and the next block could be
C-D+, D-E+, B-C+, and A-B+. Each pair
was presented twelve times for a total of 48
trials.
Probe Testing
Once criterion was met during phase
five, probe tests were given. Probe tests
consisted of the presentation of B-D+ and
X-Y+. The pair X-Y+ consisted of two
novel scents which were used to control for
141
Explorations | Social Sciences
learning the novel pairs. Twenty-four trials
of both pairs were presented each session
for three sessions.
Following B-D+ and X-Y+ tests, retraining on phase five was given. Once the rat
met criterion, more probe tests were given.
End anchors A and E were presented (AE+) and two novel stimuli were also presented (W-Z+). Twenty-four trials of both
pairs were presented each session for three
sessions.
RESULTS
Manual Procedure
S16 met baseline training criterion in 90
sessions. Figure 3 shows a breakdown of
the number of sessions per phase.
After meeting criterion for List One
Training, S16 was given probe testing with
every possible combination of stimuli.
Table 1 shows the number of trials correct for each pair for S16 and two rats from
Jordan’s (2009) study, P39 and P23. All
rats performed at above 90% correct for all
transitive pairs. S16’s total percent correct
out of the 24 trials, including baseline trials
but excluding the two trials with unfamiliar
scents, was 100%. On the two presentations of B-D+, S16 scored 2/2 (100%);
that is, he chose D over B both times. The
number of correct trials with the unfamiliar scents (X-Y+) is also presented in Table
1. S16 performed better on the tests with
transitive pairs compared to the tests with
unfamiliar scents (1/2, 50% correct).
S16 also met criterion in all four phases
during List Two training. The total number of training sessions was 59. Figure 4
shows a breakdown of the number of sessions per phase.
Table 2 shows the number of trials correct for each pair during probe testing
for S16 and two rats from Jordan’s (2009)
study, P39 and P23. All rats performed at
above 90% correct for all transitive pairs.
S16’s total percent correct, including baseline trials but excluding the two trials with
142
Figure 3. This graph shows the number of sessions per
phase in the manual procedure for List One.
Figure 4. This graph shows the number of sessions per
phase in the manual procedure for List Two.
Figure 5. This graph shows the number of sessions per
phase in the automated procedure for each subject, Y9,
Y10, Y19, and Y20.
unfamiliar scents, was 90.91%. On the
two presentations of G-I+, S16 scored 2/2
(100%). The number of correct trials with
the unfamiliar scents (W-Z+) is also presented in Table 2. S16 performed better
on the tests with transitive pairs compared
to the tests with unfamiliar scents (0/2, 0%
correct).
Automated Procedure
None of the rats trained with the automated procedure met criterion during
baseline training. The mean number of
sessions completed before training was terminated and subjects were either dropped
Mary Beth Pacewicz
from the study or given probe tests was
82.25. Figure 5 shows a breakdown of
the number of sessions per phase each rat
completed.
Y19 was dropped from the study using
the automated procedure because of a failure to meet criterion in phase two.
Y9 was presented with the A-E+ and
W-Z+ probe test in three days of probe
testing. A binomial test was conducted for
each day for percent correct on A-E+ trials
and W-Z+ trials. However, the binomial
test for choice of E on A-E+ trials was not
significant on any of the three days (15/24,
15/24, 18/24, respectively, all p>0.05).
The binomial test for choice of Z on W-Z+
trials on was not significant on days one
and two (13/24 and 16/24, p>0.05). On
day three, Y9’s performance for choice of
Z on W-Z+ trials was significant (17/24,
p<0.05). Figure 6 shows the percent correct on the final day of baseline training
and all three days of probe testing. Y9
shows little evidence of transitive inference;
while performance on the A-E+ trial finally
reached significance on day three, it was
not striking. Overall, learning appeared to
be similar to learning the relation between
the unfamiliar scents, W and Z.
Y10 was also presented with the A-E+
and W-Z+ probe test with three days of
probe testing. Again, the binomial test
for choice of E on A-E+ trials was not significant on day one (16/24, p>0.05), but
was on day two (17/24, p<0.05) and three
(19/24, p<0.01). However, the binomial
test for choice of Z on W-Z+ trials was significant on all three days (17/24, 17/24,
and 22/24, all p<0.05). Figure 7 shows the
percent correct on the final day of baseline
training and all three days of probe testing.
Y10’s improvement in performance across
the three days of probe testing provides
some evidence for learning, however, no
evidence for transitive inference. His performance was the same for A-E+ and W-Z+
trials.
Y10 was also presented with the B-D+
Figure 6. Y9’s percent correct on the final day of baseline
training and on each day of probe testing.
Figure 8. Y10’s percent correct on the final day of
baseline training and on each day of probe testing.
Figure 7. Y10’s percent correct on the final day of
baseline training and on each day of probe testing.
Figure 9. Y20’s percent correct on the final day of
baseline training and on each day of probe testing.
143
Explorations | Social Sciences
and X-Y+ probe test over three days. A binomial test was conducted for each day for
percent correct on B-D+ trials and X-Y+
trials. The binomial test for choice of D
on B-D+ trials on day one was not significant (13/24, p>0.05), but was on days
two (18/24, p<0.01) and three (21/24,
p<0.01). The binomial test for choice of Y
on X-Y+ trials on day one was not significant (14/24, p>0.05), but was on day two
(19/24, p<0.01) and three (21/24, p<0.01).
Figure 8 shows the percent correct on the
final day of baseline training and all three
days of probe testing. Y10 shows evidence
for learning over days. However, performance was identical for B-D+ and X-Y+
trials, so there is no evidence for transitive
inference.
Y20 was presented with the B-D+ and
X-Y+ probe test for three days. Figure 9
shows the percent correct on the final day
of baseline training and all three days of
probe testing. The binomial test for choice
of D on B-D+ trials was not significant on
day one (13/24, p>0.05) or two (16/24,
p>0.05), but was on day three (17/24,
p<0.05). The binomial test for choice of
Y on X-Y+ trials was not significant on any
of the three days (14/24, 15/24, 16/24,
144
all p>0.05). Thus, Y20 showed no clear
evidence of transitive inference or learning
across the probe trials.
DISCUSSION
Manual Procedure
Overall, the results from the manual
procedure show that S16 was able to make
transitive inferences with both five-item
lists. When presented with the novel pair
B-D+, S16 chose D over B both times.
Also, when presented with the novel pair
G-I+, S16 chose I over G both times. The
relations between novel pairs that emerged
indicate that a hierarchy had been formed
for both List One and List Two.
Automated Procedure
The results from the automated procedure were not very promising. Two of
the four rats in the automated procedure
never reached phase five, and of the two
rats that reached phase five, only one was
close to meeting criterion in phase five.
Performance on the transitive inference
probes was at chance levels.
Due to Y20’s health, it was unclear how
much longer he would be able to test, so he
was given the B-D+ probe test before he met
training criterion. His poor performance
Mary Beth Pacewicz
during training and on the probe tests may
be attributed to old age. Y9 was presented
with A-E+ probe tests because he reached
phase five but could not meet criterion. Y9
had achieved 10/12 on each pair, just not
on the same day. Y9 had been trained in all
five phases, and we believed he would have
been successful at the A-E+ pair. The A-E
+
pair contains two end stimuli, and therefore, should be the easiest of all novel pairs
to respond correctly to because the choice
can be made based on reinforcement histories alone (Weaver et al., 1997). Y10 was
the only rat to be tested on both the A-E+
and B-D+ pairs. He did not meet criterion
in phase five. The decision to test Y10 on
the B-D+ pair was made because he had
met criterion of 10/12 on each pair on several days, just never consecutively.
The rats’ poor performances on novel
probes in the automated procedure may be
attributed to their failure to meet criterion
during training. Other studies that involve
abstract concepts, such as Schusterman
and Kastak (1993), have shown that animals require extensive training before
probe testing. Schusterman and Kastak
demonstrated that a California sea lion
was capable of forming equivalence relations, but only after extensive training. The
California sea lion was extensively trained
with multiple exemplars to criterion before novel tests of transitivity, symmetry,
and equivalence were given. The success
on the equivalence task may be due to the
extensive training which prepared the sea
lion for the probe tests. The rats in our
automated procedure failed to fully meet
the criterion for probe testing. Therefore,
it may be argued that they did not receive
the extensive training needed to form the
five-item hierarchy.
Though the rats in the automated procedure did not reach criterion, they received much more training than the rats
in the manual procedure. In the manual
procedure, S16, was trained with 24 trials
per session for 90 sessions. Rats in the automated procedure were trained with 48
trials per session for an average of 82.25
sessions. Despite the increased amount of
training, the rats in the automated procedure did not meet criterion to receive probe
tests, whereas S16 met criterion. This
discrepancy between increased amount
of training in the automated procedure
and failure to meet criterion suggests that
amount of training does not account for
the automated procedure’s failure. By examining some of the differences between
the manual and automated procedure, possible errors and improvements for future
training in the automated procedure can
be discussed.
Comparing the Two Procedures
One difference between the manual and
automated procedure was the spatial difference between response and reinforcement. Rats trained in the manual procedure received reinforcement immediately
after making a response. Rats in the automated procedure were required to make
a response on the left side of the chamber,
but the reinforcement was delivered on
the right side of the chamber. The spatial distance between where the response
was made and reinforcement was received
also increased the time between making a
response and receiving reinforcement. In
Otto and Eichenbaum’s (1992) study using an automated apparatus to test delayed
non-match-to-sample tasks, there was not
a long time or spatial delay between response and reinforcement. Access to water
was used as reinforcement, and the water
port was located directly above the nose
poke. After making a correct response, the
rats’response was immediately reinforced.
Otto and Eichenbaum’s success in training
rats on delayed non-match-to-sample tasks
using an automated apparatus may be attributed to the minimal time between when
rats made a response and when they were
reinforced. Moving the pellet dispenser
145
Explorations | Social Sciences
between the nose pokes in the automated
procedure would decrease the spatial distance and time between when the rat
makes a response and when reinforcement
is delivered.
Using a photo beam may have also affected the rats’performances in automated
apparatus. Rather than having to make an
explicit response, such as pushing a lid in
the manual procedure, rats were required
to make a less salient response. The photo
beam was not visible to the rats, nor could
disrupting it be heard or felt. If the rats
had difficulty learning what response was
required, this may have lengthened the
amount of training needed.
The manual procedure also trained rats
with different scents than the automated
procedure. The manual procedure used
spices whereas the automated procedure
used scented oils, and all of the oils used
were fruity (e.g., strawberry, grape, and banana). Though to us the oils were distinguishable, we do not know how the scents
smelled to the rats. If this procedure was
repeated, it would be useful to use scents
that are more different (e.g., pecan, butter,
and champagne). In addition, it may be
useful to standardize the types of scents
(fruits vs. spices) across procedures. This
was not done in the current study because
of the limited variety offered by the manufacturers we used.
Ecological Significance
Another difference between the manual
procedure and the automated procedure
is the ecological significance of the required response. The importance of utilizing ecologically significant procedures
was demonstrated in a study by Wright
and Delius (1994). In this study, pigeons
were trained on match and non-match-tosample tasks using different gravel types as
stimuli. The stimuli were chosen by observing pigeons’ natural behavior of digging to find food. The pigeons learned
the task at a much quicker rate than when
146
traditional key-peck training methods are
used. Traditional key-peck training with
visual stimuli requires pigeons to peck at
lights on a screen. The key-peck method is
relatively “two-dimensional”, not allowing
extensivemanipulationwiththescreen,but
Wright and Delius (1994) found that by using three-dimensional stimuli that could be
manipulated, the pigeons learned the task
much quicker. These findings are similar
to the findings in our transitive inference
study. The manual apparatus that used
three-dimensional scented lids was much
more successful and faster at training rats
to form a five-item hierarchy.
Delius (1992) trained pigeons to discriminate between spheres and non-spheres.
The stimuli were three-dimensional objects made of various materials and textures. Acquisition was very fast compared
tootherproceduresusingtwo-dimensional
stimuli. Once the pigeons were trained,
they were presented with novel stimuli and
were able to discriminate between novel
spheres and novel non-spheres. Pictures of
the objects were also tested, and though the
pigeons did discriminate between spheres
and non-spheres, performance was lower
than performance with the three-dimensional objects. The results from Delius
(1992) suggest that the pigeons learned to
discriminate between three-dimensional
object more quickly because the stimuli
were more similar to stimuli found in the
pigeons’ natural environment. The pigeons’ high performance and fast acquisition demonstrates the importance of using
stimuli that are relevant to animals’ environmental experiences. It is possible to explain our results in terms of three-dimensional and two-dimensional stimuli. In the
manual procedure, the scents were on the
three dimensional objects, Plexiglas lids. In
the automated apparatus, the stimuli could
be considered two-dimensional, like the
pictures in Delius (1992). In order to make
a response in the automated apparatus, the
Mary Beth Pacewicz
rats did not have to manipulate the stimuli.
Delius found that acquisition and performance were higher with three-dimensional
stimuli compared to two-dimensional stimuli. These results are similar to our findings that the manual procedure succeeded
in training a five-item hierarchy, but the
automated procedure required more training and was not successful in training a
hierarchy.
If the automated procedure was more
difficult for the rats to learn because it is not
ecologically significant, rats in this procedure may require more extensive training.
Training the rats with different programs
may help them learn the hierarchy more
quickly. For example, Davis (1992) trained
rats to form a five-item hierarchy by training one pair at a time. In phase one, only
the A-B pair was presented. In phase two,
only the B-C pair was presented. Phase
three trained rats on the A-C pair, and
phase four only presented the C-D pair.
Phase five presented the D-E pair, phase
six trained the C-E pair, and phase seven
trained the A-E pair. The rats received
extensive training on each adjacent pair as
well as nonadjacent pairs. This procedure
may have been successful because the rats
were given experience with nonadjacent
pairs before the transitive inference test.
Similarly, before the test of equivalence
in Schusterman and Kastak‘s (1993) study,
the California sea lion was given multiple
exemplars and had been trained and
tested on tests or symmetry, transitivity, and
equivalence. By first training the sea lion
with symmetry, transitivity, and equivalence, the sea lion was familiar with these
types of tests. Familiarizing our rats with
the type of test that will be given on a probe
day may be a useful addition to the automated procedure. Our automated procedure could be modified to train rats in a
similar way as Lazareva and Wasserman
(2006) and Davis (1992). Familiarizing our
rats with the type of test that will be given
on a probe day may be a useful addition to
the automated procedure.
The wide range of species (e.g., macaque
monkeys (Treichler & Van Tilburg, 1996),
pigeons (von Ferson, Wynne, Delius, &
Staddon, 1991; Lazareva & Wasserman,
2006), and rats (Davis, 1992; Dusek &
Eichenbaum, 1997)) that have succeeded
on transitive inference tasks may be related to abilities required for living in social groups. Animals living in social groups
must be able to recognize others in their
groups as well as monitor social rank and
status. For example, male elephant seals
fight for access to females and less than
one-third of them actually mate each season (LeBouef, 1974). Body size is one of
the factors that influences each males social
rank. Central males are usually the largest
and most experienced fighters. Peripheral
males are restricted to the periphery of the
social group, and outside males have little,
if any access to females (Modig, 1996).
Instead of fighting every male in the group,
a newcomer can determine his place in the
hierarchy based on his interactions with
just a few individuals.
Male elephant seals may utilize physical characteristics of group members to
form a hierarchy. However, hyenas are
an example of a species that does not use
physical dimensions when monitoring social rank. The social status of male hyenas
is not based on physical characteristics,
but is determined by the hyena’s “place in
line.” Males essentially wait to rise in status
rather than fight each other. Male hyenas
that join a new group rarely use physical
means to secure status. Rather, they are
made aware of their status by a greeting
ceremony. During the greeting ceremony,
the new male is approached by another
male of similar social status. By evaluating the male’s relationship with all of the
other members of the group, the new male
is able to determine his social status (Slater,
Rosenblatt, Snowdon, & Roper, 2002).
147
Explorations | Social Sciences
The need to monitor and evaluate social
status based on relationship to other group
members requires the formation of hierarchies. The formation of hierarchies in
novel situations allows an animal to determine rank without fighting several group
members. Rodents such as meadow voles,
African striped mice, and rats have complex
social interactions that result in the formation of dominance hierarchies (Spritzer,
Meikle, & Solomon, 2004; Kinahan &
Pillay, 2008; Calhoun, 1963). This capability is an adaptive strategy that requires the
use of properties of transitive inference.
Conclusion
Thus, it is important to continue studying
148
transitive inference in laboratory settings.
While our results do not provide evidence
that an automated apparatus is an effective
tool for training rats in transitive inference
tasks, Jordan’s (2009) success in training
rats in two five-item lists suggests that using
a manual procedure is a preferred training
method. However, using an automated apparatus has several advantages compared
to using a manual procedure. For example,
objectivity and efficiency are increased.
Therefore, improvements to the automated
apparatus should be made before it is determined that the automated procedure
should not be used in training rats in transitive inference tasks.
Mary Beth Pacewicz
REFERENCES
Calhoun, J. B. (1963). The Ecology and Sociobiology of the Norway Rat. Bethesda, Maryland:
US Department of Health, Education, & Welfare.
Davis, H. (1992). Transitive inference in rats (Rattus norvegicus). Journal of Comparative
Psychology, 106(4), 342-349.
Delius, J. D. (1992). Categorical discrimination of objects and pictures by pigeons. Animal
Learning and Behavior, 20(3), 301-311.
Dusek, J. A., & Eichenbaum, H. (1997). The hippocampus and memory for orderly
stimulus relations. Proc. Natl. Acad. Sci., 94, 7109-7114.
Gillan, D. J. (1981). Reasoning in the chimpanzee: II. Transitive inference. Journal of
Experimental Psychology, 7(2), 150-164.
Hayes, S. C. (1989). Nonhumans have not yet shown stimulus equivalence. Journal of the
Experimental Analysis of Behavior, 51(3), 385-392.
Jordan, L. (2009). List linking as a tool to test transitive inference in rats. Explorations: The
JournalofUndergraduateResearchandCreativeActivitiesfortheStateofNorthCarolina,4,
85-107.
Kinahan, A. A., & Pillay, N. (2008). Dominance status influences female reproductive
strategy in a territorial African rodent Rhabdomys pumilio. Behavioral Ecology and
Sociobiology, 62, 579-587.
Lazareva, O. F., Smirnova, A. A., Bagozkaja, M. S., Zorina, Z. A., Rayevsky, V. V., &
Wasserman, E. A. (2004). Transitive responding in hooded crows requires linearly ordered stimuli. Journal of the Experimental Analysis of Behavior, 82, 1-19.
Lazareva, O. F., & Wasserman, E. A. (2006). Effects of stimulus orderability and reinforcement history on transitive responding in pigeons. Behavioural Processes, 72,
161-172.
Lazareva, O. F., & Wasserman, E. A. (2008). Categories and concepts in animals. Learning
and Memory: A Comprehensive Reference, 1, 197-226.
LeBouef, B. J. (1974). Male-male competition and reproductive success in elephant seals.
Amer. Zool., 14(1), 163-176.
Lionello-DeNolf, K. M., & Mihalick, S. M. (2006). An automated olfactory discrimination procedure with rats. The Behavior Analyst Today, 7(4), 560-571.
McIntire, K. D., Cleary, J., & Thompson, T. (1987). Conditional relations by monkeys:
Reflexivity, symmetry, and transitivity. Journal of the Experimental Analysis of
Behavior, 47(3), 279-285.
149
Explorations | Social Sciences
Modig, A. O. (1996). Effects of body size and harem size on male reproductive behavior
in the southern elephant seal. Animal Behaviour, 51(6), 1295-1306.
Otto, T., & Eichenbaum, H. (1992). Complementary roles of the orbital prefrontal cortex
and the perirhinal-entorhinal cortices in an odor-guided delayed-nonmatchingto-sample task. Behavioral Neuroscience, 106(5), 762-775.
Schusterman, R. J., & Kastak, D. (1993). A California sea lion (Zalophus californianus) is
capable of forming equivalence relations. The Psychological Record, 43, 823-839.
Sidman, M. (2000). Equivalence relations and the reinforcement contingency. Journal of
the Experimental Analysis of Behavior, 74(1), 127-146.
Sidman, M., Rauzin, R., Lazar, R., Cunningham, S., Tailby, W., & Carrigan, P. (1982).
A search for symmetry in the conditional discriminations of rhesus monkeys,
baboons,andchildren.JournaloftheExperimentalAnalysisofBehavior,37(1),23-44.
Slater, P. J. B., Rosenblatt, J. S., Snowdon, C. T., Roper, T. J. (2002). Advances in the study of
behavior. San Diego, CA: Academic Press.
Spritzer, M. D., Meikle, D. B., & Solomon, N. G. (2004). Journal of Comparative Psychology,
118(3), 332-339.
Treichler, F. R., & Van Tilburg, D. (1996). Concurrent conditional discrimination tests
of transitive inference by macaque monkeys: list linking. Journal of Experimental
Psychology, 22(1), 105-117.
Von Fersen, L., Wynne, C. D. L., Delius, J. D., & Staddon, J. E. R. (1991). Transitive
inferenceformationinpigeons.JournalofExperimentalPsychology,17(3),334-341.
Weaver, J. E., Steirn, J. N., Zentall, T. R. (1997). Transitive inference in pigeons: control
for differential value transfer. Pyschonomic Bulletin & Review, 4(1), 113-117.
Wright, A. A., & Delius, J. D. (1994). Scratch and Match: pigeons learn matching and
oddity with gravel stimuli. Journal of Experimental Psychology, 20(1), 108-112.
150